Unable to create custom array codecs

[JoseLion]

Hi,

I'm using spring-data-r2dbc and the extension mechanism from r2dbc-postgresql to create a codec for a custom Postgres range type:

CREATE TYPE timetzrange AS RANGE (subtype = timetz);

Since there's no native support for Postgres range types on Java's side, I created my own class OffsetTimeRange, and registered a custom codec OffsetTimeRangeCodec which implements the interface Codec<OffsetTimeRange>.

So far, that worked perfectly fine, but I wanted to add the custom codec for its corresponding multirange type. From Postgres range types documentation, we know that:

When you define your own range you automatically get a corresponding multirange type.

So I registered another codec OffsetTimeRangeArrayCodec which implements the interface Codec<OffsetTimeRange[]>, but this raised the following exception upon reading/writing from the database:

java.lang.IllegalArgumentException: Unsupported array type: com.example.util.OffsetTimeRange

The problem

I did some debugging, and I found the exception is thrown in the following check (line #164): https://github.com/spring-projects/spring-data-relational/blob/b691af7b305adfb0a4f8ea7585dc8f7c0abc80e1/spring-data-r2dbc/src/main/java/org/springframework/data/r2dbc/dialect/PostgresDialect.java#L156-L168

The problem seems to be that the this.simpleTypeHolder.isSimpleType(typeToUse) method checks against the local simple types, which are missing any type introduced by a custom codec.

As always, I'll be happy to help with a PR if someone can point me in the right direction 🙂

Cheers!

[mp911de]

Spring Data has an abstraction to detect which types are simple ones (understood natively by the driver). For Spring Data R2DBC, we introduced a Dialect abstraction to keep database-specifics within a single type. A dialect object can also tell which Java types are simple types via the R2dbcDialect.getSimpleTypes() method.

Subclassing PostgresDialect, enriching simple types, and configuring the dialect in your application should allow you to use your application-specific simple types.

[JoseLion]

@mp911de thanks, that worked! However, it was not that easy to get there 😅. Is this just a workaround, or is it documented somewhere?

Also, it'd be great if there could be a simpler way to add custom array types. In a nutshell, these are all the steps I followed to make it work:

• Create a custom codec implementing Codec<OffsetTimeRange[]>
• Add the codec to the CodecRegistrar
• Create a custom @WritingConverter instance:
  • Implementing Converter<OffsetTimeRange, OffsetTimeRange>, where the convert(source) method returns the same source value
• Add the custom writing converter to the R2dbcCustomConversions bean
• Create a CustomProstgresDialect class that extends from PostgresDialect
  • Override the getSimpleTypes() method
  • Add OffsetTimeRange.class to the collection without losing the types from super.getSimpleTypes()
• Create a CustomDialectProvider class that implements R2dbcDialectProvider
  • Use the getDialect(..) method to return a CustomProstgresDialect instance
• Finally, as mentioned in the docs, create the file META-INF/spring.factories that should contain the line:

org.springframework.data.r2dbc.dialect.DialectResolver$R2dbcDialectProvider=com.example.CustomDialectProvider

As you can see, there are a lot of steps and new files. Is there any way to improve this process?

As always, I'll be happy to contribute if possible 🙂

[mp911de]

The CodecRegistrar is part of the ConnectionFactory configuration and we cannot do much here. The WritingConverter shouldn't be necessary. Enums use that mechanism only to bypass the built-in enum conversion.

R2dbcDialectProvider should also not be required as you can override either AbstractR2dbcConfiguration.getDialect() when using Spring Data R2DBC without Spring Boot. With Spring Boot, you currently cannot provide a custom dialect, but that would be something for Spring Boot to improve. You could alternatively provide your own R2dbcCustomConversions bean as R2dbcCustomConversions can be created from a dialect and a collection of converters.

[JoseLion]

Yep, the CodecRegistrar it's easy to use (a great feature, IMO!). However, with Spring Boot I still need the WritingConverter, or else I get the following exception:

org.springframework.dao.InvalidDataAccessApiUsageException: Nested entities are not supported

But... the error goes away if I override AbstractR2dbcConfiguration.getDialect() with my custom dialect

On Spring Boot, I was able to create a @Configuration class that extends from AbstractR2dbcConfiguration to make some of those overrides. I had to create the ConnectionFactory manually (using the PostgresqlConnectionFactoryProvider builder), which is not a bad thing because it also gave me easy access to register custom codecs 😁

I tried providing my own R2dbcCustomConversions bean created with my custom dialect, but it didn't work. It looks like the check for the simple types uses a dialect provided before the one from the R2dbcCustomConversions bean. That's why I had to switch to the META-INF/spring.factories solution 😅

However, I think the most straightforward solution on Spring Boot was to have a @Configuration class that extends from AbstractR2dbcConfiguration, so we can easily add more customizations. Would this be something ok to do on Spring Boot?

Thanks again for your answers, @mp911de! They have been of great help 🙂